??Summer Vibes: Your AI in Coaching Beach Read

In this summer newsletter edition we cover your essential AI Beach Reads ?? :

• Latest research on AI and Theory of Mind
• A story on AI Native schools
• AI developers news relevant for coaches and educators
• Our own Kou?ing centar updates

Can AI do Theory of Mind?

Jill had just moved into a new house. She went shopping with her Mum and bought some new curtains. When Jill had just put them up, her best friend Lisa came round and said, "Oh, those curtains are horrible, I hope you're going to get some new ones." Jill asked, "Do you like the rest of my bedroom?"

In the story, did somebody say something they should not have said?

What the person said that they should not have said?

Did Lisa know that the curtains were new?

The example above is one of the tasks used in psychological research to examine Theory of Mind. What is Theory of Mind?

It is an ability to understand and track other people’s mental states, like their beliefs, desires, intentions, and emotions. This ability is essential for effective social interactions and communication.

In coaching, Theory of Mind allows coaches to empathize with clients, anticipate their needs, and respond appropriately to their emotional and psychological states. It underpins the coach’s ability to provide personalized guidance and support, making it a cornerstone of effective coaching practice.

Now, the key question: Can AI do Theory of Mind? The answer to this was explored in a recent paper titled Testing theory of mind in large language models and humans, published in Nature Human Behavior. The short answer: yes, AI (GPT and Llama2) can perform or even outperform human participants across the battery of theory of mind tests.

On a more nuanced level of analysis, the paper revealed some of the specifics of AI when it comes to "faux pas" tasks (like the one in the introductory example above). For example. Llama2 showed a bias towards attributing ignorance: when faced with a situation where a character might or might not know a crucial piece of information, this AI model often assumed the character did not know it. While this bias helped it excel in detecting faux pas—since faux pas often involves a speaker unknowingly making an inappropriate remark—it might not reflect a genuine understanding of social nuances but rather a heuristic shortcut. GPT-4, on the other hand, displayed a hyperconservative approach towards committing to conclusions: it was cautious in making definitive statements about a character's mental state unless it was very clear from the context. While this conservatism might reduce the risk of errors in some cases, it also led to underperformance in tasks like faux pas detection, where inferring unspoken social context is crucial. This approach likely stems from the model's design to avoid over-committing to potentially incorrect inferences.

As authors conclude, "these findings not only demonstrate that LLMs exhibit behaviour that is consistent with the outputs of mentalistic inference in humans but also highlight the importance of systematic testing to ensure a non-superficial comparison between human and artificial intelligences."

These implications touch upon International Coaching Federation recent draft of AI Coaching Standards, in which AI coaching systems are expected to have the capability to do emotional state assessment and response in a manner that is empathetic and appropriate to client's emotional needs (Element B.5.3).

The ICF document underscores the significant role emotions play in AI coaching, particularly in how sentiment analysis can enhance coaching effectiveness. Connecting this to the findings of the Nature article, we see that Llama2 and GPT-4's differing biases in emotional attribution can significantly impact AI coaching outcomes. Llama2's tendency to attribute ignorance might lead to a more conservative approach, ensuring clients do not feel overwhelmed, while GPT-4's cautiousness in reaching conclusions can help maintain a balanced emotional state, preventing overconfidence. These AI behaviors align with the ICF's emphasis on a balanced emotional experience, facilitating optimal learning and growth in clients.

Share your thoughts on AI and Theory of Mind in the comments below!

AI Native schools: Are we ready

Andrej Karpathy , former Head of AI at Tesla and researcher at OpenAI, has launched Eureka Labs , an innovative "AI native" education platform. The key idea behind Karpathy's new project is to transform education by integrating AI teaching assistants into traditional learning environments. Eureka Labs envisions a future where AI assistants support human teachers, making high-quality education accessible to everyone.

The startup's initial offering, LLM101n, is envisioned as an undergraduate-level class which teaches students to train their own AI models. The course materials will be available online, with both digital and physical cohorts to enhance collaborative learning. Eureka Labs' approach aligns with the potential benefits highlighted in studies, such as improved student grades through AI teaching assistants. By combining human expertise with AI's scalability, Eureka Labs aims to expand educational reach and depth, transforming how subjects are taught and learned.

The launch of Eureka Labs by Andrej Karpathy, with its focus on AI native education, offers exciting implications for the coaching education industry. Here are some of the ideas for integrating AI agents in coaching education: scaling, personalization, continuous improvement, data insights.

At Kou?ing centar along with our high touch human expertise in coaching education, we are developing AI agents for coaching education. Our project Coachia is all about that - AI learning system for human skills development and its debut is our ICF accredited program in Coaching Education, starting September 4th (write to us on [email protected] to learn more about this exciting innovation in the field of coaching education). Can we call it AI native Coaching Education? Share your thoughts in the comments below.

Anthropic's Artifacts launched

In this newsletter we have previously written about Anthropic and Claude AI. We have also analyzed Claude Opus capabilities related to coaching tasks in our own research paper. This July, Anthropic has been releasing many useful innovations related to its latest model Claude 3.5 Sonet. One of the features that caught our attention is Artifacts. Here is how the model itself explains the feature.

It is interesting to see how this promise of long-term, project-based collaboration plays out for coaches, L&D experts and organizational leaders.

With our vision for Coachia to make it a model-agnosti AI learning system, we are definitely going to give a try with Claude's Artifacts feature. What we will be looking with Artifacts is more streamlined educational content management, longer memory across sessions to ensure continuity and maintain context without repetitive explanations and even more effective collaboration between our learners and educators. Will report about progress in some of the future newsletter editions!

Open source watch: Nomic AI & Moshi chat

Open source AI means democratizing access and supporting innovation. It's on our roadmap and we are monitoring closely its developments.Well, OpenAI made a big announcement with Voice mode somewhere in May, but we are still waiting to try talking to AI as showcased at the official demo event. French AI developer Kyutai introduced something similar - called Moshi .

Moshi is designed to provide conversations with users through voice. According to Kyutai, Moshi can speak in various accents and has 70 different emotional and speaking styles. The AI can even handle two audio streams simultaneously, allowing Moshi to listen and talk simultaneously.

I couldn't resist and tried out Moshi for a bit. Here's what I got.

Not bad in my opinion. We are so super inpatient to see voice as part of our AI learning system, that any news related to this really thrills us! Multimodality, with voice in the focus, is all about immersion and natural conversational flow in L&D. Just imagine you needn't type any longer to communicate with AI!

Moshi, keep up with the good work. We'll be coming back for more.

Another open source release that caught our attention in July is Nomic AI , that launched GPT4All 3.0. What it does is lets you access thousand of LLMs locally on your computer. Thers's been a stream of launches around locally running AI with Copilot+PC and Apple Intelligence.

With Nomic's release it's now an open source project that focuses on privacy and security, making it a popular choice both for individual users and enterprises.

For coaches and L&D professionals, ethics, privacy and security are key. Continuous development of open source AI platforms makes us closer to fully making data privacy a solved issue once for goods.

Health related behavior change with AI: Thrive.ai

Sam Altman, CEO of OpenAI, and Arianna Huffington, CEO of Thrive Global, collaboratively wrote an article in Times on the potential of AI for behavior change related to health.

The paper is a sort of a lunch pad for the Thrive AI Health initiative, funded by OpenAI Startup Fund and Thrive Global , with an aim to create a personalized AI health coach. This coach, integrated with personal biometric data and behavior change methodologies, promises to provide real-time, tailored health recommendations.

Altman & Huffington discuss how AI can enhance daily health behaviors—sleep, diet, exercise, stress management, and social connection. This AI health coach could help busy professionals manage conditions like diabetes by offering personalized reminders and suggestions. Unlike generic health recommendations, this coach will offer precise, actionable advice, such as meal swaps or exercise prompts, democratizing health benefits and addressing inequalities. AI could thus transform health care, promoting both physical and mental well-being through improved daily habits.

We are looking at these developments with great attention and care as they pave the way to coaching innovations with AI into the spheres of health related behaviors. Something like a new era of personal relation to health.

Share your comments below!

Kou?ing centar July AI story

We are also furthering our research related to AI in coaching, this time via our AI in Coaching: Stakeholders' Attitudes Survey. We promised to publish prelim results in this newsletter, so here they are for the first 2 weeks.

Prelim results reveal from certified coaches, followed by AI enthusiasts, tech professionals and HR professionals, reveal a generally positive sentiment towards the use of AI in coaching and coach training. Majority trust AI tools less than traditional methods. Respondents have noted significant improvements in the quality, efficiency, and scalability of coaching sessions with AI. Interestingly, AI Simulator is perceived as the most beneficial AI learning agent. Self perceived knowledge of AI tools in coaching is rated as moderate or low, while the education in AI is seen as driving positive attitudes towards its application in coaching and coach training.

Help us gather more perspectives by sharing our survey!

Our July story is about demoing Coachia. Reach out to us for a 1:1 demo or simply join our demo waiting list.

And a little bit of positive vibes from us for the end. We are delighted to announce that Kou?ing centar has been accepted into the Microsoft for Startups program! This partnership is a significant milestone for us as we continue to innovate and lead in the field of AI-augmented organizational development and learning. Joining the Founders Hub provides us with invaluable resources and support, propelling our mission to transform learning experiences.

This opportunity will enhance our upcoming AI-augmented ICF certification program, set to launch in September, and strengthens our commitment to delivering unparalleled value to our clients.

Thank you to Microsoft for this incredible opportunity. We are excited about the journey ahead and the impact we will create together.